Classifying flows and buffer state for YouTube’s HTTP adaptive streaming service in mobile networks Dimitrios Tsilimantos, Theodoros Karagkioules, and Stefan Valentin Mathematical and Algorithmic Sciences Lab, Paris Research Center Huawei Technologies France fdimitrios.tsilimantos, theodoros.karagkioules,
[email protected] Abstract—Accurate cross-layer information is very useful to and play-back buffer state of that stream in real time. This optimize mobile networks for specific applications. However, information allows schedulers, traffic shaping and admission providing application-layer information to lower protocol layers control schemes to minimize their impact on Quality of has become very difficult due to the wide adoption of end-to- end encryption and due to the absence of cross-layer signaling Experience (QoE) or to even increase it by providing bit-rate standards. As an alternative, this paper presents a traffic profiling guarantees when possible [7]. solution to passively estimate parameters of HTTP Adaptive This demand for accurate application-layer information is Streaming (HAS) applications at the lower layers. By observing a major practical problem. Network optimization functions IP packet arrivals, our machine learning system identifies video typically operate at the Layer 2 and 3 of the ISO/OSI protocol flows and detects the state of an HAS client’s play-back buffer in real time. Our experiments with YouTube’s mobile client show stack, while application information is available at Layer 7. that Random Forests achieve very high accuracy even with a Currently, MNOs solve this cross-layer signaling problem by strong variation of link quality. Since this high performance a combination of explicit signaling or Deep Packet Inspection is achieved at IP level with a small, generic feature set, our (DPI).